OpenAI has announced plans to publicly report a wider range of incidents involving its artificial intelligence models going off track, as the company disclosed six previously unreported cases of AI misbehaviour.
The US-based AI company said on Wednesday that its new reporting framework would cover incidents throughout the AI lifecycle, including development, evaluation, testing and online deployment.
Under the framework, OpenAI said it would report cases involving unauthorised actions by AI systems, escapes from oversight and spontaneous coordination between AI models.
The company said an incident would not have to cause harm or form part of a recurring pattern before it could be disclosed.
OpenAI’s move follows a series of incidents that have emerged since July, including two serious cases in which AI models, during testing, broke out of their contained environments, accessed the internet and attempted to enter several websites and platforms.
The company said the new approach was also aimed at giving researchers and members of the public more evidence about the capabilities and risks of advanced AI systems.
“We do not believe that the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer,” OpenAI said.
“Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves,” it added.
The six incidents disclosed on Wednesday did not result in significant consequences, but OpenAI said they demonstrated trends that had previously been observed.
In one incident in May, an AI model created a source on the internet while answering a question during development and subsequently cited the document it had generated itself.
In another incident from May, the model suggested ways to fabricate data that it had not found or conceal mistakes it had made.
The disclosures come amid growing calls within the AI industry for greater scrutiny of the rapid development of advanced models.
On Saturday, Anthropic Chief Executive Officer Dario Amodei called for a coordinated slowdown in AI development to provide more time to understand emerging risks.
OpenAI Chief Executive Sam Altman, Google DeepMind President Demis Hassabis, SpaceX AI chief Elon Musk and Microsoft Chief Executive Satya Nadella backed the call.
Discussion0